Network reconfiguration is of theoretical and practical significance to guarantee safe and economical operation of distribution system. In this paper, based on all spanning trees of undirected graph, a novel genetic algorithm for electric distribution network reconfiguration is proposed. Above all, all spanning trees of simplified graph of distribution network are found. Tie branches are obtained with spanning tree subtracted from simplified graph. There is one and only one switch open on each tie branch. Decimal identity number of open switch on each tie branch is taken as the optimization variable. Therefore, length of chromosome is very short. Each spanning tree corresponds to one subpopulation. Gene operations of each subpopulation are implemented with parallel computing method. Individuals of offspring after gene operation automatically meet with radial and connected constraints for distribution network operation. Disadvantages of conventional genetic algorithm for network reconfiguration that a large amount of unfeasible solutions are created after crossover and mutation, which result in very low searching efficiency, are completely overcome. High calculation speed and superior capability of the proposed method are validated by two test cases.
Batch-Constrained Reinforcement Learning for Dynamic Distribution Network Reconfiguration
